The Contribution of Educational Changes to Fertility Declines in Low‐ and Middle‐Income Countries

Abstract

Educational expansion is widely regarded as a central mechanism in fertility transitions, yet some research suggests its role may be more modest. This study analyzes birth histories from 414 surveys in 72 low‐ and middle‐income countries to reassess the contribution of the shift in the educational composition of the population to fertility declines. We reconstruct fertility trends by educational level over several decades and apply the Kitagawa decomposition method to determine whether declines primarily reflect shifts in educational composition or changes in fertility within educational groups. On average, educational expansion accounts for approximately 30 percent of the decline in fertility, with substantial variation across regions, countries, and over time. The declines primarily reflect fertility changes within educational groups, especially among women without secondary education, rather than shifts in the educational composition of the population. However, although educational progress is neither necessary nor sufficient for fertility decline, it has consistently contributed to fertility reductions and is likely to remain an important component of fertility transitions.

BACKGROUND

Over the past decades, changes in educational attainment in low‐ and middle‐income countries (LMICs) have been notable and transformative. Increased access to education and improvements in educational systems have been central to these changes, resulting in higher enrollment rates (World Bank ²⁰¹⁸). Many countries have made substantial progress toward universal primary education (United Nations General Assembly ^{2000, 2015}). Furthermore, the gender gap in education has narrowed significantly, with more girls gaining access to schooling and pursuing higher levels of education (Duflo 2012). These changes have had positive implications for economic development, as education is an important driver of poverty reduction and economic growth in these countries (Psacharopoulos and Patrinos 2018).

Education is also a key determinant of fertility in LMICs (Castro Martín ¹⁹⁹⁵; Cleland ²⁰⁰²; Schultz ²⁰⁰²), operating through three main paths. First, education contributes to postponing childbearing by delaying the onset of sexual activity and marriage (Bongaarts, Mensch, and Blanc ²⁰¹⁷; Chemhaka and Odimegwu ²⁰¹⁹; Esteve and Florez‐Paredes ²⁰¹⁸; Hertrich ²⁰¹⁷). Second, women with higher educational attainment often exhibit reduced desire for larger families (Hertrich 2017; Shapiro ²⁰¹²; Shapiro and Tenikue ²⁰¹⁷). Third, better educated women tend to adopt contraceptives more frequently, leading to a decline in unplanned pregnancies and births (Bongaarts, Mensch, and Blanc ²⁰¹⁷; Casterline ²⁰¹⁷; Liu and Raftery ²⁰²⁰). Beyond these proximate mechanisms, education may enhance women's autonomy, increase labor market participation and the opportunity costs of childbearing, and shift norms surrounding family size and investments in children (Liu and Raftery 2020; Schoumaker ²⁰¹⁷).

While a vast literature and numerous survey results demonstrate the (often large) fertility differentials by educational level in LMICs (Bongaarts 2003; Diamond, Newby, and Varle ^1999; Goujon, Lutz, and KC 2015), the role of increasing levels of education in fertility decline is less clear‐cut. Some authors mention the rise in education levels is among “the key causes of fertility decline during the demographic transition” (James, Skirbekk, and Van Bavel ²⁰¹², 1). Yet, several pieces of research suggest that “schooling of the adult population may be a relatively modest force behind fertility decline” (Cleland 2002, 194).

When assessing the link between educational expansion and fertility trends at the country level, one can differentiate between two key dimensions (Bongaarts, Mensch, and Blanc ²⁰¹⁷; Cleland ²⁰⁰²; Eloundou‐Enyegue, Giroux, and Tenikue ²⁰¹⁷; Schoumaker and Sánchez‐Páez ²⁰²⁴). First, as educational attainment rises, shifts in the educational composition of the population will contribute to lower fertility rates (the composition effect): women with higher levels of education typically have lower fertility and their fertility influences the country‐level total fertility rate (TFR) more heavily as educational attainment rises. Second, fertility can change within specific educational groups (the rate effects) and influence aggregate fertility trends, even without changes in the educational composition of the population. In practice, both composition and rate effects can simultaneously contribute to aggregate fertility trends.

Decomposition methods disentangle these effects on fertility change. They are not designed to identify the causal effect of education on fertility. Instead, they quantify the extent to which fertility declines reflect shifts in the educational composition of the population versus changes in fertility behavior within educational groups. Large composition effects suggest that educational expansion, particularly for girls, is an important component of fertility reduction. However, the presence of a large compositional effect does not imply that education alone drove the decline, as other structural changes, such as urbanization, may occur alongside educational expansion. In contrast, when rate effects clearly dominate, this pattern suggests that other factors—including family planning programs—contribute more substantially to fertility reduction, with declines occurring largely independently of educational expansion.

Previous decomposition analyses examining the relationship between educational expansion and fertility decline have highlighted substantial cross‐national and temporal variation in the relative importance of composition and rate effects. Early work by Weinberger, Lloyd, and Blanc (1989) on four Latin American countries in the 1970s and 1980s showed that although changes in educational composition accounted for a substantial part of the fertility decline, rate effects (changes within educational groups) were larger in three of the four countries studied. Subsequent research by Castro Martín (1995), covering seven countries across Africa, Latin America, and the Caribbean, confirmed this pattern. The study found that rate effects were larger than composition effects in all cases except for Peru. It also emphasized that composition effects were not uniform across contexts and accounted for a larger share of fertility change in Latin American and Caribbean countries than in African countries.

Analyses conducted in the early 2000s reinforced these conclusions. Cleland (2002) examined changes in TFRs between the 1970s and 1990s in eight countries across North Africa, sub‐Saharan Africa, Asia, and Latin America. This study identified that within‐group fertility changes (rate effects) were “the dominant force of change” (Cleland 2002, 194) and that composition effects exceeded 50 percent of total fertility change only in Peru and the Philippines. In contrast, composition effects were especially small in Bangladesh, Morocco, and Kenya. Around the same time, Bongaarts (2003) analyzed the links between fertility and education in 57 less developed countries and 10 European countries. While the study did not formally use decomposition methods, it argued that “much of the overall decline in fertility [was] attributable to declines of fertility within each group rather than to an upward shift in the distribution of levels of schooling” (Bongaarts 2003, 324). However, Bongaarts also noted that if substantial educational fertility differentials persisted in the last stages of the fertility transition—as it appeared to be the case in low fertility countries—the shift in the educational composition of the population would be “a key determinant of a country's overall level of fertility” (Bongaarts 2003, 333).

More recent evidence from Latin America supports the predominance of rate effects, while also highlighting substantial cross‐country variation. An analysis of fertility changes in five Latin American countries between 2010 and 2022 found that approximately one‐fifth of total fertility change was attributable to composition effects, with shares ranging from 2 percent in Chile to 37 percent in Costa Rica (Cabella and Velázquez 2025). This study shows that, even in the later stages of the fertility transition and in countries where substantial educational fertility differentials remain, within‐group fertility changes can still be the dominant factor in aggregate fertility decreases. Overall, existing evidence indicates that aggregate fertility change has more frequently resulted from within‐group fertility changes rather than from shifts in the educational composition of women. It also indicates substantial variation across countries and over time.

This paper extends prior research on the links between educational expansion and fertility decline in LMICs, examining whether aggregate fertility declines are primarily attributable to shifts in the educational composition of the population or to fertility changes within educational groups. Our study makes four contributions. First, our analysis includes a substantially larger set of LMICs and surveys, spanning a longer period and incorporating the most recent data. Previous decomposition studies typically examined only four to eight countries over short intervals (around 10 years). In contrast, we analyze 72 countries, several of which have been observed over more than 40 years. This is made possible by exploiting full birth histories to reconstruct long‐term fertility trends. This coverage also enables examination of the links between educational changes and fertility declines across a variety of settings and at different fertility levels, and the identification of distinct regional patterns. Second, we harmonize the classification of educational groups across countries. Earlier studies have employed different definitions of educational groups, which may impede cross‐country and over‐time comparisons. Third, for selected countries, we compare results using four educational categories. Previous research has usually used a three‐level categorization of education. While we also mainly use a three‐level approach, we assess the sensitivity of our findings to a finer categorization in countries where a substantial number of women have attained higher education. Fourth, our decomposition goes beyond estimating overall composition and rate effects by quantifying the contribution of fertility change within each educational group to aggregate fertility decline.

We show that the contribution of educational expansion to fertility decline (the composition effect) has varied substantially across countries and over time. In some countries, the composition effect is large; in others, it is limited, reflecting differences in both the pace of educational expansion and the magnitude of educational fertility differentials. As in previous research, we find that educational expansion is not the dominant force in global fertility declines, but it nevertheless accounts for a substantial portion of the decline (roughly 30 percent). Fertility changes within educational groups—reflecting the impact of other factors—account for the largest share of global fertility changes, especially when fertility has not reached low levels. We also identify regional differences, with educational expansion accounting for a smaller share of fertility decline in Asia and North Africa than in sub‐Saharan Africa, Latin America, and the Caribbean. Moreover, the relative importance of composition effects increases as countries approach low fertility levels.

DATA AND METHODS

Data Sources

We use 414 Demographic and Health Surveys (DHS), Multiple Indicator Cluster Surveys (MICS), World Fertility Surveys (WFS), and Reproductive Health Surveys (RHS) from 72 LMICs. These surveys collect birth histories and sociodemographic characteristics (including education) among large representative samples of women aged 15–49. The list of countries and surveys is available in Table A1 (in the Supporting Information).¹

Education was harmonized into three levels: (1) no education, (2) 1–6 years of education (corresponding to some or completed primary education in most countries), and (3) seven or more years of education (some secondary education and over). This harmonized classification facilitates comparisons over time in countries where the education system changed during the studied period (as in Ghana and Nigeria) and across countries with different education systems (Schoumaker and Sánchez‐Páez ²⁰²⁴). The three‐level categorization is a compromise between a limited number of categories that ensures sufficient observations in each group across all countries and periods, and a more detailed classification that better reflects heterogeneity in educational attainment and fertility differentials across educational groups. However, in some countries, the share of women with higher education (13 or more years of schooling) has increased substantially over time. Relying on a broad category of women with seven or more years of education may understate the composition effect, particularly at later stages of the fertility transition. Therefore, in selected countries, we also assess the sensitivity of the results to a four‐category classification (no education, 1–6 years, 7–12 years, and 13 or more years).

Analyses of Microdata

We reconstruct fertility trends by educational level across all 72 countries for the longest available period by combining several surveys. To this end, for each survey, we created tables of births and exposure by calendar year (for 15 years preceding the survey), five‐year age groups, and levels of education.² For each country, the number of births and exposures by age group, calendar year, and education are summed across all surveys. For each educational group, a Poisson regression is used to estimate TFRs by five‐year age groups and periods, with these variables treated as independent categorical variables. We assume that the age pattern of fertility is constant within each survey but can vary over time (Schoumaker 2014). Pooling birth histories in this way enables the reconstruction of fertility trends over long periods (e.g., longer than 50 years in some countries, such as Ghana, Kenya, and Colombia), while limiting random fluctuations and the effects of data quality issues.

We compute fertility rates by level of education measured at the time of the survey.³ This approach assumes that educational attainment at the time of the survey approximates women's level of education at age 15 and can be treated as time invariant (Hoem and Kreyenfeld 2006). In some countries, a substantial proportion of women enrolled in school at age 15 or 16 (and more rarely at older ages) may still be in primary education and transition to secondary school later. If they have reached secondary school by the time of the survey, they are classified as having secondary education for the entire 15–19 age group, even if they were attending primary school for part of that period. However, because fertility rates at age 15 and 16 are low in most countries, the impact on TFRs will be limited.⁴

The same data are used to reconstruct trends in the share of women by educational level. We use the exposure (person‐years) in each educational group among women aged 15–39 for the 15 years preceding each survey. The exposures by calendar year and educational group are summed across all surveys. For each five‐year period, the share of women by educational level is computed as the ratio of women's exposure in the educational group to the total exposure. The combination of retrospective estimates of education from several surveys reduces random variations and the effects of data quality issues. It leads to smoother trends than those obtained with survey‐specific estimates.

Application of the Decomposition Method

We use the Kitagawa decomposition method to quantify the portion of fertility change attributable to shifts in the population's educational composition (Kitagawa 1955; Preston, Heuveline, and Guillot ²⁰⁰¹; Schoumaker and Sánchez‐Páez ²⁰²⁴). The decomposition of the changes in the TFR between two dates (t1 and t2) is expressed as follows (Equation 1):

$$\begin{array}{ccc}\hfill TF{R}_T^{t2}-TF{R}_T^{t1}& =& \left(TF{R}_S^{t2}-TF{R}_S^{t1}\right)\times \left[\frac{P_S^{t2}+P_S^{t1}}{2}\right]+\left(TF{R}_P^{t2}-TF{R}_P^{t1}\right)\times \left[\frac{P_P^{t2}+P_P^{t1}}{2}\right]\hfill \\ & & +\left(TF{R}_N^{t2}-TF{R}_N^{t1}\right)\times \left[\frac{P_N^{t2}+P_N^{t1}}{2}\right]\hfill \\ & & +\sum _{i=1}^K\left(P_i^{t2}-P_i^{t1}\right)\times \left[\frac{TF{R}_i^{t2}+TF{R}_i^{t1}}{2}\right]+Residual\hfill \end{array}$$ (1)

The left‐hand side of the equation is the aggregate fertility change (at the country level) between time 1 (t1) and time 2 (t2). The first product on the right‐hand side is the fertility change among women with 7+ years of education (subscript S), weighted by the average share of women (ages 15–39) with 7+ years of education. The second product is computed similarly for women with 1–6 years of education (subscript P), and the third product for women with no education (subscript N). These three products measure the contribution of fertility changes within each educational group to the country‐level change in fertility. Their sum is the rate effect. The fourth term captures the contribution of changes in educational composition to overall fertility change (the composition effect). It is a weighted sum of changes in the proportion of women 15–39 (P) in each educational group, with weights equal to the group's average fertility level. The residual accounts for the fact that the observed country‐level TFR differs slightly from the weighted average of education‐specific TFRs, with weights given by the shares of women aged 15–39 by educational level. This is usually very small and has a minor influence on the results. The decomposition method can be used for successive five‐year estimates or longer periods, for example, over the whole period covered by the surveys.⁵ The ratio of the composition effect to the total fertility change expresses the percentage of the overall change attributable to shifts in the educational distribution. This ratio will usually lie between 0 and 100 percent. However, in some cases, it may fall outside this range, for example, when fertility increases overall, or when a negative composition effect more than offsets a positive rate effect.

This decomposition method is a descriptive tool that allows the difference between two rates to be allocated into components that reflect changes in the population's composition and within‐group changes (Kitagawa 1955, 1172). When the composition effect dominates, fertility would have declined more slowly in the absence of educational expansion. Conversely, when within‐group fertility change (the rate effect) accounts for most of the decline, aggregate trends are likely driven primarily by factors other than educational expansion.

Accounting for Uncertainty

Our estimates are subject to sampling errors as the analysis relies on survey data. We therefore compute confidence intervals using a stratified cluster bootstrap because analytical standard errors are not available for the decomposition. This procedure also accounts for the two‐stage sampling design and within‐cluster correlation. For each country, we perform 500 bootstrap replications, resampling primary sampling units with replacement while maintaining the original number of clusters and treating surveys as strata. Confidence intervals are derived from the empirical distribution of the bootstrap estimates.

We present 95 percent confidence intervals for TFRs and the proportions of women by educational level in the figures showing fertility and education trends. We also report 95 percent confidence intervals for all the components of the decompositions by broad periods in Table A2 (in the Supporting Information). Their magnitude is acceptable in most countries. For example, for the composition effect expressed as a percentage of the total fertility change, the median width of confidence intervals is around 8 percentage points. In countries with very large samples and strong fertility declines, the confidence intervals are considerably smaller (e.g., less than 2 percentage points in India and in Bangladesh). Estimates are less precise in countries with small sample sizes and/or limited fertility changes, where confidence intervals can exceed 50 percentage points. Such situations are rare, occurring only in a few countries, such as the Congo and Mauritania.

Illustration of the Method in Haiti and Bangladesh

The following two examples illustrate the method and the diversity of results that can be expected across countries and over time. Figure 1a shows the reconstructed five‐year TFR estimates by educational level in Haiti. Figure 1b shows trends in the proportion of women by educational level in the same country, and Figures 1c and 1d present the decomposition results by broad periods and by five‐year periods. Figures 2a–2d illustrate these results for Bangladesh.

FIGURE 1.

Changes in fertility and education, and decomposition of fertility changes in Haiti

SOURCE: World Fertility Survey and Demographic and Health Surveys.

FIGURE 2.

Changes in fertility and education, and decomposition of fertility changes in Bangladesh

SOURCE: Demographic and Health Surveys and Multiple Indicator Cluster Surveys.

In Haiti, fertility had already begun to decline at the national level (black line in Figure 1a) in the 1960s. The pace of decline was moderate between the early 1960s and the late 1970s (0.05–0.10 children per year). Fertility then increased slightly between the late 1970s and late 1980s, and the decline resumed at a faster pace in the 1990s. Fertility among women with 7+ years of education (green line) was already much lower in the 1960s than among women with primary education (dark blue line) and no education (light blue line). While fertility changed within educational groups, these within‐group changes have been limited, and the fertility differentials by level of education have remained very large over time, consistent with the “permanent difference model” (Bongaarts 2003). Over the same period, educational levels increased substantially. Approximately 60 percent of the women (15–39) had attained 7+ years of schooling in the mid‐2010s, compared with only 5 percent in the early 1960s. The changing composition of the population by educational levels contributed to lower fertility. This is illustrated by the fact that the TFR declined much more rapidly at the country level than within educational groups.

The decomposition method quantifies the contributions of shifts in educational composition and within‐educational‐group fertility changes to overall fertility changes. Figure 1c shows the contribution of each element over the entire period. Between the mid‐1960s and the mid‐2010s, fertility declined by 0.062 children per year on average. The gray areas represent the composition effect, while the colored areas represent the contribution of within‐group fertility changes to the aggregate change in fertility. The combination of large fertility differentials across educational groups and the shift in the distribution of women by education means that a large share—61 percent (95 percent CI: 53–69 percent)—of Haiti's fertility decline over the 50‐year period is associated with changes in educational composition. In contrast, fertility changes within educational groups (mainly among women with no education and with 1–6 years of education) have contributed more modestly to the overall fertility decline (Figure 1c). Figure 1d, which depicts changes across successive five‐year periods, also shows that rate and composition effects have varied over time. Rate effects were substantial in some periods—either increasing or decreasing fertility—but were not the dominant factor in the decline over the whole period.

The Bangladesh case (Figures 2a–d) illustrates a markedly different situation. Between the 1980s and the 2020s, fertility declined at a sustained pace (−0.10 children per year on average). Over the 40‐year period, educational changes accounted for only a limited share of the fertility decline—16 percent (95 percent CI: 15–17 percent)—and most of the decline results from within‐group fertility changes. This reflects the fact that women's education progressed slowly until the mid‐1990s, whereas fertility changes within educational groups were rapid during the same period. When secondary education expanded faster from the early 2000s, educational‐level fertility differentials had already become small, limiting the contribution of educational changes to fertility declines. Fertility changes within educational groups account for 84 percent of the total changes (Figure 2c), with the largest contributions from women with no schooling—44 percent (95 percent CI: 42–46 percent)—and women with 1–6 years of education—29 percent (95 percent CI: 28–31 percent). Figure 2d clearly shows that the rapid decline in the 1980s and early 1990s was primarily due to a decline in fertility among less educated women.

These two countries illustrate that shifts in educational composition can be a minor contributor to fertility decline (as in Bangladesh), a major factor (as in Haiti), or a modest yet significant force in countries with intermediate situations. These two examples also show that composition effects vary across the fertility transition: they can be substantial in some periods, and more limited in others.

Aggregation of Country‐Level Findings to the Regional and Global Levels

In the Results section, we present findings for the 72 countries (Figures A1–A4 in the Supporting Information) and summarize patterns at the regional and global levels. Stylized trends in fertility and education are estimated using mixed‐effects models with country‐specific random intercepts. Trends over time by educational group, within each region and at the global level, are modeled as smooth functions of calendar year using natural splines with two degrees of freedom. In other words, for each region and educational group, a single average time trend is estimated, while allowing country‐specific intercepts to vary. These figures are intended to illustrate broad regional patterns rather than to provide precise regional level estimates of fertility and educational trends. Comparisons of rate and composition effects at the global and regional levels, and across different fertility levels, are presented using boxplots. The boxplots display the median, the interquartile range, whiskers extending to 1.5 times the interquartile range, and outliers (observations beyond the whiskers).

RESULTS

Figures A1–A4 (in the Supporting Information) present fertility trends by educational level across 72 countries, trends in the share of women by level of education, and decompositions of changes for the entire period and for five‐year periods.⁶ We summarize the main results below.

Fertility Differentials and Changes by Education

Figure 3 illustrates fertility trends by educational level in three broad groups of LMICs (Asia and North Africa, sub‐Saharan Africa, and Latin America and the Caribbean), as well as across all countries combined. Watermarked lines represent country‐specific estimates, and thick lines indicate the average fertility trends by educational level across the three groups of countries and for all countries combined, obtained with the mixed‐effect models. The regional averages show patterns that are also visible in most individual countries. First, fertility is lower among better educated women (those with seven or more years of education) in almost all countries and across most periods of the fertility transition. Small fertility differences by educational level are found in only five countries (Bangladesh, Egypt, Indonesia, Morocco, and Tunisia), and only in recent periods of low fertility (Figure A1 in the Supporting Information). In all 67 other countries, women with seven or more years of education have fewer children, even when the fertility transition is nearly complete. This is evident at the global level and in each broad region, where the fertility rates of women with seven or more years of education are systematically lower than those of women with lower levels of education. Women with some primary education also tend to have lower fertility than women with no education. Still, differences are smaller, especially at the beginning and the end of the fertility transition. Regional differences also indicate that fertility differentials by level of education are smaller in Asia and North Africa (especially in the most recent period) and larger in Latin America and the Caribbean. As discussed before, substantial fertility differentials are necessary for shifts in educational composition to affect aggregate fertility trends, and composition effects are expected to be larger in Latin America and the Caribbean than in Asia and North Africa.

FIGURE 3.

Changes in fertility by educational levels in three broad regions and the 72 countries—stylized trends

NOTES: Thick lines show averages obtained from a random‐effects model with natural splines (2 knots).

SOURCE: Demographic and Health Surveys, Multiple Indicator Cluster Surveys, Reproductive Health Surveys, World Fertility Surveys.

Second, regional averages indicate a clear decline in fertility across all educational groups (Figure 3), a finding also visible in many countries (Figure A1 in the Supporting Information). Even though differences across educational groups persist over time in most countries, the most common pattern is one in which fertility also declines among women with no schooling. The declines among less educated women have been especially pronounced in Asia and North Africa (e.g., in Morocco and Bangladesh), where fertility differentials have become very small. They have also been substantial among less educated women in Latin America and the Caribbean (e.g., in Peru and Ecuador) but remain at higher levels. In sub‐Saharan Africa, fertility changes among women with no education have been more modest but have been slightly faster than among the better educated women in recent years. The fact that fertility tends to decrease across all educational groups means that rate effects will account for part of the overall decline in fertility.

Robust Educational Progress at Varying Paces

All countries have also experienced educational progress, characterized by an increase in the share of women with seven or more years of education and a decrease in the share of women without education (Figure 4). However, educational changes have been uneven across countries and over time, and educational levels still vary enormously. On average, progress has occurred earlier in Latin America and the Caribbean. Educational changes have also been significant in Asia and North Africa, but slower in sub‐Saharan Africa. In some countries (e.g., Colombia, the Dominican Republic, Indonesia, Zimbabwe), secondary education attainment is very high (close to or exceeding 90 percent), and a small proportion of women remain out of school. In other countries (e.g., Burkina Faso, Ethiopia, Niger), approximately half of women still have no education, and fewer than a quarter have seven or more years of education. Overall, changes have been regular, except in a few countries (e.g., Madagascar in the late 1990s–early 2000s), which experienced slowdowns or reversals in educational progress.

FIGURE 4.

Educational changes in three broad regions and the 72 countries—stylized trends

NOTES: Thick lines show averages obtained from a random‐effects model with natural splines (2 knots).

SOURCE: Demographic and Health Surveys, Multiple Indicator Cluster Surveys, Reproductive Health Surveys, World Fertility Surveys.

Ultimately, the variety of educational changes, combined with educational fertility differentials, should lead to diverse composition effects across countries and over time. The population share of each educational group also influences its contribution to aggregate fertility change. For instance, rapid fertility declines among better educated women will have limited influence at the national level if they constitute only a small share of the population. In contrast, even modest fertility changes among women with no education can make a substantial contribution to the decline in countries with low educational attainment. We address these questions with the decomposition method.

Educational Changes Account for around 30 Percent of Fertility Declines

The median fertility decline across 72 countries is equal to −0.070 children per year (one child per 14 years), with the fastest decline in Latin America and the Caribbean (−0.090 children per year), the slowest in sub‐Saharan Africa (−0.060 children per year), and intermediate values in Asia and North Africa (−0.073 children per year).

The decomposition of fertility changes over broad periods (Figure A3 in the Supporting Information) shows that the contribution of education to fertility changes has varied substantially across countries. These periods vary in length and rarely cover the entire fertility transition. Yet, the decomposition provides orders‐of‐magnitude estimates for the composition and rate effects. Figure 5 summarizes the composition and rate effects across broad regions using boxplots. Overall, the median composition effect (in gray) in the 72 countries is −0.021 children per year, equivalent to 1 child per 47 years (Figure 5). It is the largest in Latin America and the Caribbean (median value of −0.028 children per year, 1 child per 36 years), where fertility differentials by educational groups are large, and educational progress has been strong. Smaller composition effects are observed in Asia and North Africa (−0.019 children per year, 1 child per 53 years), where fertility differentials by education are much smaller and progress in education has been less rapid than in Latin America and the Caribbean. The median value is also −0.019 children per year in sub‐Saharan Africa, reflecting slower educational progress than in the other regions.

FIGURE 5.

Contribution of rate effects and composition effects in fertility declines in three broad regions and the 72 countries

SOURCE: Demographic and Health Surveys, Multiple Indicator Cluster Surveys, Reproductive Health Surveys, World Fertility Surveys.

Figure 5 also shows rate effects. On average, they have been much larger than the composition effects, showing the important contribution of fertility changes within educational groups to declines in country‐level fertility. The median rate effect across the 72 countries (−0.049 children per year, 1 child per 20 years) is more than twice the size of the median composition effect. It is the largest in Latin America and the Caribbean (−0.062 children per year, 1 child per 16 years), followed by Asia and North Africa (−0.054 children per year, 1 child per 19 years), and lowest in sub‐Saharan Africa (−0.041 children per year, 1 child per 24 years). Rate effects also tend to be more variable than composition effects, reflecting the less predictable trends in fertility than in educational progress. In some countries (e.g., Angola and the Democratic Republic of Congo), changes in fertility within educational groups have contributed to increases in fertility. In contrast, composition effects have contributed to decreases in fertility across all countries.

To estimate the relative contribution of composition effects to fertility declines, we divide the median composition effect by the median change in total fertility. At the global level (across 72 countries), composition effects account for 30 percent of the decline in fertility. This ratio is a little higher in Latin America and the Caribbean (31 percent) and in sub‐Saharan Africa (32 percent). In contrast, in Asia and North Africa, composition effects account for only 26 percent of the changes, once again reflecting the smaller fertility differentials across educational groups and the substantial fertility changes within these groups.⁷ These broad regional patterns also mask substantial variations across countries. In each region, composition effects have contributed significantly to fertility declines in some countries (e.g., more than 50 percent in Mozambique, Haiti, and Turkey), while they were much lower in others (e.g., less than 25 percent in Malawi, Belize, and Bangladesh). They exceed 50 percent in only seven of the 69 countries in which fertility declined (Table A2 in the Supporting Information). In contrast, rate effects predominate in 50 countries and indeterminate situations (confidence intervals that include the 50 percent value) are found in 12 countries.

Fertility Changes within Educational Groups

The rate effects, that is, changes within educational groups, account for the largest share of the global decline in fertility (around 70 percent). Have fertility declines been driven primarily by fertility changes within one group or have all groups contributed roughly equally? Figure 6 distinguishes the three components of the rate effects, one for each educational group. On average across the 72 countries, women with 1–6 years of education contributed slightly more (−0.014 children per year) than women with at least secondary education (−0.011) or women with no education (−0.009). Differences across groups are much more pronounced in Latin America and the Caribbean, where women with primary education have made the largest contribution to fertility declines (−0.042 children per year), far larger than the contributions of women with secondary education (−0.017), and with no education (−0.009). In Asia and North Africa, women with no education have contributed more (−0.022) than the more educated groups (−0.019 among women with 1–6 years of education, −0.009 among women with 7+ years of education). Finally, the contributions are more similar across the three groups in sub‐Saharan Africa (−0.007 children per year among women with no education, −0.008 among women with primary education, and −0.010 among the better educated women). Again, we find strong heterogeneity across countries. For instance, women with no education have had by far the largest contribution to fertility declines in some countries (e.g., Burkina Faso, Guinea, Morocco), resulting from the large share of these groups in the population and the substantial fertility declines within these groups; women with secondary education were the largest contributors to the declines in a few countries (e.g., Eswatini, Jordan, Lesotho), and women with some primary education in a few others (e.g., Indonesia, Honduras). Overall, these findings show substantial variation in the relative contribution of each educational group to changes in total fertility.

FIGURE 6.

Contribution of rate effects by educational groups and composition effects in fertility declines in three broad regions and the 72 countries

SOURCE: Demographic and Health Surveys, Multiple Indicator Cluster Surveys, Reproductive Health Surveys, World Fertility Surveys.

Variations of Composition Effects across Fertility Levels

The previous analyses were performed using the decomposition of changes by broad periods. We now assess whether the composition effects differ between high‐, intermediate‐, and low‐fertility contexts. We summarize the results of the five‐year decomposition analyses (Figure A4 in the Supporting Information) and classify each period by the TFR at the beginning of the interval, which serves as a rough indicator of the stage of the fertility transition. For each five‐year period, we measure the composition effect, the rate effect (the sum of the three components), and their median values and distributions across fertility levels (Figure 7).

FIGURE 7.

Contribution of rate effects and composition effects in fertility declines by stages of the transition in the 72 countries

SOURCE: Demographic and Health Surveys, Multiple Indicator Cluster Surveys, Reproductive Health Surveys, World Fertility Surveys.

We find that educational expansion contributes to fertility declines at all fertility levels, including in high‐fertility contexts with more than six children per woman. Composition effects are slightly larger at intermediate levels (fertility between 4.5 and 6.0 children) and lowest at low‐fertility levels. On average, composition effects are markedly smaller than rate effects, except at low fertility levels, where the two effects are small and comparable in magnitude. We also find that rate effects are more variable than composition effects, which almost invariably contribute to fertility declines.

Sensitivity of Results to Alternative Education Groupings

Educational expansion implies that an increasing share of women within the broad 7+ years category has attained higher levels of education. Failing to account for this growing proportion of highly educated women may lead to attributing part of the observed fertility decline to rate effects, when it in fact reflects compositional change within this broad category. This issue was discussed by Bongaarts (2003, 322), who argued that “the broadness of the educational categories is unlikely to have a significant effect on the main conclusions of [his] study.” However, given that in an increasing number of countries more than 20 percent of women have attained higher education, this issue warrants closer empirical scrutiny.

We assess the sensitivity of the results to a four‐level classification in a subset of 15 countries (six from Latin America and the Caribbean, three from sub‐Saharan Africa, six from Asia and North Africa) in which the share of women with 13+ years of education had reached at least 20 percent in the most recent survey and where fertility decreased during the studied period. Country‐specific figures (fertility trends, educational trends, decomposition) for this subset of countries are presented in Figures A5–A8 (in the Supporting Information), and the relative composition effect (percentage of total fertility change) is compared across the three‐level and four‐level categorizations (Table A3 in the Supporting Information).

In most of these countries, the fertility of women with 13+ years of education is substantially lower than that of their less educated counterparts, although differences have been small in recent periods in some countries (e.g., Tunisia, Egypt, Vietnam).⁸ Overall, composition effects account for a slightly larger share of the fertility decline in these 15 countries with the four‐level grouping (median of 33 percent vs. 31 percent with the three‐level grouping), but the general conclusion remains unchanged. However, in a limited number of countries, the use of a four‐level classification matters substantially. For instance, in Vietnam, it nearly doubles the composition effect (from 9 percent to 16 percent) relative to a three‐level grouping. In contrast, in a few countries, such as Egypt and Bolivia, composition effects are virtually the same with a four‐level classification as with a three‐level classification. Overall, the four‐level classification does not substantially affect the results, but this may become an increasingly important consideration as educational levels rise.

CONCLUSION

This article's objective was to reconstruct fertility trends by educational level, document educational expansion, and quantify the contribution of educational expansion to fertility decline in 72 LMICs. It shows that female educational expansion has occurred everywhere. Although the changes occurred at different paces across countries, they were substantial overall. Women with at least some secondary education were scarce in the 1960s and 1970s, but they now represent the largest group in most countries. Conversely, the share of women with no education has declined markedly worldwide, although this group remains non‐negligible in sub‐Saharan Africa. We also find that educational‐level fertility differentials are substantial in many countries and across most of the fertility transition, especially for women with seven years or more of education. Despite these regularities, our analyses of fertility trends by educational level also highlight substantial heterogeneity in fertility differentials across educational levels, as well as a variety of patterns of fertility change within educational groups. The decomposition method was used to quantify the portion of fertility change attributable to educational expansion and to within‐group fertility changes.

Our results confirm some findings from earlier studies, while qualifying others. First, the decomposition indicates that shifts in educational attainment are not the primary components in fertility changes. As Bongaarts (2003) noted over two decades ago, a larger portion of fertility declines appears to be associated with changes in fertility behavior within educational groups rather than shifts in the distribution of women across these groups. Yet we estimate that approximately 30 percent of the aggregate decline in fertility is attributable to shifts in the educational distribution over time. While educational expansion is not the dominant component, it is far from negligible.

Our median estimate of the contribution of educational expansion to fertility declines, based on 414 surveys in 72 countries, is comparable to estimates from earlier studies using fewer surveys and shorter time periods. For instance, the median estimate for the seven countries in Castro Martín's (1995) study was 33 percent. In Cleland's (2002) analysis of eight countries in the 1970s and 1980s, it was 24 percent. Weinberger, Lloyd, and Blanc (1989) estimated a higher median of 44 percent for four Latin American countries, whereas more recent estimates for five Latin American countries indicated a lower contribution (median estimate of 24 percent) from composition effects (Cabella and Velázquez 2025). Taken together, these 24 previous estimates yield a median contribution of 32 percent, very close to our global estimate of 30 percent. Conversely, the decomposition shows that changes in fertility rates within educational groups account for the largest share of fertility declines in most countries.

As in previous research, we also find considerable variation in situations across countries and regions. In some countries, shifts in the distribution of women across educational strata account for a large share of fertility declines, whereas in many others their contribution is more limited, and in a few it is negligible. Overall, there is no single pattern regarding the link between changes in education and fertility declines, but we find some regional differences that could not be inferred from previous studies. For instance, the contribution of educational expansion has been more limited in Asia and North Africa than in other regions, whereas rate effects (fertility changes within educational groups) have been larger in these countries. This was especially clear in countries such as Bangladesh, Cambodia, and Indonesia, where fertility declined rapidly across all educational groups and where educational fertility differentials were small in recent periods. Such situations correspond to cases in which significant downward fertility trends occurred, or would have occurred, without substantial changes in education (Basu 2002), which may indicate the impact of strong family planning programs (Castro Martín ¹⁹⁹⁵). These patterns suggest that rapid fertility decline can occur even in the absence of large compositional shifts in education, particularly where effective family planning programs and other factors influence reproductive behavior across all educational groups. In such contexts, rate effects dominate.

We also find that the contributions of composition effects and rate effects vary with the level of fertility. At high levels of fertility (i.e., in the first phases of the fertility transition), fertility changes within educational groups (rate effect) tend to contribute more to the overall decline in fertility than educational progress (composition effect). At lower fertility levels, the composition effect plays a larger role relative to the total effect. This is consistent with Bongaarts’ (2003, 333) hypothesis that “educational composition can play a key role in shaping levels and trends in fertility at the end of the transition” in contexts where fertility differentials remain substantial. But other paths to low fertility are also possible, as demonstrated by the rapid fertility declines among less educated women and the almost converging levels of fertility across educational groups.

This study also has some limitations. As mentioned at the beginning of this article, the decomposition method does not isolate the causal effect of educational changes on fertility. The estimated “contribution” of educational expansion to fertility declines thus partly includes the influence of other structural changes, and in that sense overestimates the contribution of educational changes. Conversely, the decomposition method may underestimate the role of education in fertility decline, as educational expansion can also have indirect effects on fertility. For instance, higher educational attainment can lead to changes in fertility among less educated women through shifts in fertility norms, labor market dynamics, and other structural factors that reduce fertility regardless of educational attainment (Kravdal 2002, 2012). In other words, the fertility decline among less‐educated women (captured by rate effects) can be influenced by rising educational attainment. Another limitation is that identical educational categories (no education, 1–6 years, 7+ years) may have different meanings across space and time, for instance, due to differences in educational systems or in the quality of education (Chen 2025). We should also note that other decomposition methods can yield similar conclusions but slightly different results. Finally, as demonstrated in previous research (Bongaarts, Mensch, and Blanc ²⁰¹⁷; Frye and Lopus ²⁰²⁵), the contribution of education to fertility changes can be much stronger at the onset of fertility, and, conversely, more limited at higher ages. These limitations suggest that measuring the exact contribution of education to fertility decline is challenging, if not illusory. Nevertheless, our findings indicate that educational expansion plays a substantial but highly context‐dependent role in fertility decline, even if it is not the dominant driver. Although education is neither a necessary nor a sufficient condition for rapid fertility reduction, continued expansion of female education is likely to remain a central component of fertility change in LMICs.

CONFLICTS OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Supporting Information

DATA AVAILABILITY STATEMENT

The data supporting the findings of this study are openly available in UCLouvain's dataverse (https://doi.org/10.14428/DVN/AT2VYB) (Schoumaker and Sánchez‐Páez ²⁰²⁶). The dataset includes estimates of education‐specific TFRs and the distribution of women aged 15–39 by educational level, computed from survey data. The survey microdata used in this study are available from the WFS, DHS, MICS, and RHS archives. Access to these data may be subject to restrictions.